A Novel Visualization Model for Web Search Results

1
A Novel Visualization Model for Web Search Results

• An Application of the Solar System Metaphor
• Tien N. Nguyen and Jin Zhang
• Electrical and Computer Engineering Department
• Iowa State University
• USA

2
Motivation

• WWW becomes an important source of information.
• Internet search engines provide tremendous help
  in information seeking.
• Search engines present relevant Web pages in a
  list with titles and short paragraph descriptions
  extracted from the texts.
• Browsing the ranked list can become rather
  tedious and unproductive.

3
Motivation

• The order of keywords used in a query also
  affects the final outcomes of a search.
• information system, system information
• A word or a phrase can have multiple meanings
  depending on different contexts.
• information visualization
• Contextual information or topics of interest can
  only be introduced as a different query.
• software visualization

4
Web Search Result Viz

• Must provide an effective guide in searching for
  needed information
• Controllable and adaptable
• Abstraction of our visualization problem
• The query
• The search results
• Context information or topics of interest
• Semantic relations
• Degree of relevance

5
Solar System as Metaphor

• All planets and asteroids rotate around the sun
  in the universe.
• Each of them has its own orbit and a moving
  speed.
• They are attracted to each other by gravity.
• It is gravity that determines the orbit and
  moving speed of an asteroid.

6
Solar System as Metaphor
7
Solar System as Metaphor
(Subject)

• Location
• Distance
• Color
• Speed
• Gravity

(Page)
(Query)
8
Generalization

• Applications
• Web
• search
• Paper
• citation
• Visualization
• Hyperlink
• structure

9
WebSearchViz Space
10
Parameter Computation

• R is the radius.
• Simij is the semantic strength between two
  entities (e.g. between a page and a query, or
  between a page and a subject). It is between 0
  and 1.
• Context icons are anchored on the circumference.
  Objects are inside the circle.
• The distance of an object
• di R x (1 Simi,center)
• Objects with high Sims are close to center.

11
WebSearchViz Space
(?k, Rcons)
12
Parameter Computation

• The angle of an object against X-axis
• If Sim1i Simmi ltgt 0
• If an object that is irrelevant to any displayed
  contexts, ?i is defined as zero.
• That is, it is projected onto the X-axis.

13
Rotations and Analyses

• Activate a context icon to rotate evenly around
  the circumference.
• Assume that the moving speed of a specified
  context is ? degrees per second and its angle is
  ?r with initial value of ?0.
• The dynamic angle of an object i is

14
Rotations and Analyses

• The rotation speed of an object i is
• All affected objects (i.e. objects that are
  somewhat related to the rotating context) will
  move evenly in their orbits.
• The moving speed is affected primarily by Simri,
  the degree of relevance between object i and
  context r.

15
Rotations and Analyses

• If Simri 0 (no relevance between object i and
  context r), the object stays put.
• If an object is only relevant to the moving
  context and is not to others, it has the same
  speed as the moving context

16
Rotations and Analyses

• In this case, the object icon is always located
  on the line connecting the moving context and the
  center.

17
Project ambiguity

• Project ambiguity many objects in a
  high-dimensional space are projected into
  low-dimensional space.
• Project ambiguity and a large number of objects
  can cause visual occlusion.
• Two objects are projected onto the same points if
  their Sim value is 1.
• Overlapping icons can be disambiguated via
  changing the positions of contexts.

18
Visual Occlusion

• To handle overlapping in the case of many
  objects
• Automatic clustering/grouping functionality
• Zoom in and zoom out
• For Web search results We provide focus page
  shifting functionality
• A page can become the new center
• Surrounding objects are relevant pages.

19
WebSearchViz system
(Subject)

• Location
• Distance
• Color
• Speed

(Page)
(Query)
20
WebSearchViz architecture
21
Vector-based Representation
Keyword 1 Keyword j Keyword p
Page 1 0.1224 0.0593
Page 2 0.0349 0.0671
Page i Wij

Page n

• Wij is computed via a modified version of the
• TF-IDF algorithm
• Each Web page is represented by a row (a
  vector).
• Similarity of two pages (Sim) is computed via a
  combination of cosine and Euclidean distance
  values.
• A query or a subject is represented as a vector.

22
Subject Definition
23
WebSearchViz System
http//www.eng.iastate.edu/ tien/WebSearchViz
24
Conclusions

• Solar System provides a nice metaphor for the
  novel visualization model
• The use of movement and speed adds an extra
  dimension into a visual space, illustrating the
  semantic connections among objects.
• Web Search result visualization
• Accommodate the changes of preferences and
  subjects of interest

25
Thank you

• Tien N. Nguyen
• Electrical and Computer Engineering Department
• Iowa State University
• tien_at_iastate.edu